Update app.py

app.py

@@ -1,9 +1,16 @@
 # ================================================================
 #  ProSync AI — The Event Producer's Command Center
-#  Handles both B2B (Corporate) and B2C (Private) events.
+#  Gradio application for Hugging Face Spaces
+#
+#  Data source : HF Dataset repo  eliel2003/events  (vendors file)
+#  Embed model : sentence-transformers/all-MiniLM-L6-v2
+#  Scoring     : 60% semantic similarity + 40% composite quality
 # ================================================================
 
+import spaces   # required by HF GPU Space infrastructure — do not remove
+
 import os
+import io
 import json
 import warnings
 
@@ -12,54 +19,64 @@ import pandas as pd
 import torch
 import gradio as gr
 from sentence_transformers import SentenceTransformer, util as st_util
-import spaces
 
 warnings.filterwarnings("ignore")
-os.environ["CUDA_VISIBLE_DEVICES"]    = ""
-os.environ["TOKENIZERS_PARALLELISM"]  = "false"
+os.environ["CUDA_VISIBLE_DEVICES"]   = ""
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
-# Required by HF GPU Space infrastructure — do not remove
+# Required by HF GPU Space infrastructure — satisfies the
+# "@spaces.GPU function detected" startup check.
 @spaces.GPU
-def dummy_gpu_function():
+def _gpu_stub():
     pass
 
-# ── Config ────────────────────────────────────────────────────
+# ── Configuration ─────────────────────────────────────────────
 HF_TOKEN       = os.environ.get("HF_TOKEN", "")
+HF_DATASET     = "eliel2003/events"
 EMBED_MODEL_ID = "sentence-transformers/all-MiniLM-L6-v2"
 
-# Private events get a different (warmer, personal) prompt tone
+# ── Domain constants (match notebook exactly) ─────────────────
+ALLOC_RATIOS = {
+    "Catering":          0.304,
+    "Venue":             0.228,
+    "AV_Technology":     0.175,
+    "Entertainment":     0.104,
+    "Photography_Video": 0.076,
+    "Logistics":         0.057,
+    "Security":          0.057,
+}
+VENDOR_CATEGORIES = sorted(ALLOC_RATIOS.keys())
 
-VENDOR_CATEGORIES = [
-    "Catering", "AV_Technology", "Venue", "Security",
-    "Photography_Video", "Entertainment", "Logistics",
-]
 CATEGORY_EMOJI = {
-    "Catering": "🍽️", "AV_Technology": "🎬", "Venue": "🏛️",
-    "Security": "🛡️", "Photography_Video": "📷",
-    "Entertainment": "🎭", "Logistics": "🚚",
-}
-ALLOC_RATIOS = {
-    "Catering": 0.304, "Venue": 0.228, "AV_Technology": 0.175,
-    "Entertainment": 0.104, "Photography_Video": 0.076,
-    "Logistics": 0.057, "Security": 0.057,
+    "Catering":          "🍽️",
+    "AV_Technology":     "🎬",
+    "Venue":             "🏛️",
+    "Security":          "🛡️",
+    "Photography_Video": "📷",
+    "Entertainment":     "🎭",
+    "Logistics":         "🚚",
 }
-CITIES  = ["Beer Sheva", "Haifa", "Herzliya", "Jerusalem",
-           "Netanya", "Petah Tikva", "Ramat Gan", "Tel Aviv"]
+
+CITIES  = [
+    "Beer Sheva", "Haifa", "Herzliya", "Jerusalem",
+    "Netanya", "Petah Tikva", "Ramat Gan", "Tel Aviv",
+]
 SEASONS = ["Winter", "Spring", "Summer", "Fall"]
 EVENT_TYPES = [
-    "Annual Conference", "Award Ceremony", "Bar/Bat Mitzvah", "Brand Activation",
-    "Corporate Gala", "Family Reunion", "Investor Day", "Private Birthday",
-    "Product Launch", "Team Building", "Tech Summit", "Trade Show",
+    "Annual Conference", "Award Ceremony", "Bar/Bat Mitzvah",
+    "Brand Activation", "Corporate Gala", "Family Reunion",
+    "Investor Day", "Private Birthday", "Product Launch",
+    "Team Building", "Tech Summit", "Trade Show",
     "Wedding", "Workshop Series",
 ]
+
 QUICK_STARTERS = [
     {
         "label":  "🏙️  Tech Summit · Tel Aviv",
         "brief":  "Large-scale tech summit — advanced AV, LED walls, live streaming, "
                   "kosher catering for 400 guests, VIP executive security.",
         "city": "Tel Aviv", "season": "Summer", "budget": 250_000,
-        "type": "Tech Summit", "guests": 400,
-        "date": "October 15, 2026", "venue": "The Tel Aviv Convention Center",
+        "type": "Tech Summit", "guests": 400, "date": "2026-10-15",
         "notes": "Kosher catering required. VIP lounge for 30 executives.",
     },
     {
@@ -67,8 +84,7 @@ QUICK_STARTERS = [
         "brief":  "Elegant annual corporate gala — plated fine dining, live band, "
                   "professional photography and videography for 200 guests.",
         "city": "Jerusalem", "season": "Winter", "budget": 140_000,
-        "type": "Corporate Gala", "guests": 200,
-        "date": "December 5, 2026", "venue": "The King David Hotel Ballroom",
+        "type": "Corporate Gala", "guests": 200, "date": "2026-12-05",
         "notes": "Black-tie dress code. Award presentation segment.",
     },
     {
@@ -76,8 +92,7 @@ QUICK_STARTERS = [
         "brief":  "Outdoor team building day — interactive entertainment, DJ, "
                   "logistics, casual catering for 150 employees.",
         "city": "Haifa", "season": "Spring", "budget": 65_000,
-        "type": "Team Building", "guests": 150,
-        "date": "April 22, 2026", "venue": "Carmel Forest Resort",
+        "type": "Team Building", "guests": 150, "date": "2026-04-22",
         "notes": "Outdoor venue preferred. Vegetarian options required.",
     },
     {
@@ -85,17 +100,17 @@ QUICK_STARTERS = [
         "brief":  "Intimate outdoor wedding — elegant catering, DJ, floral design, "
                   "photography, and logistics for 250 guests.",
         "city": "Netanya", "season": "Spring", "budget": 120_000,
-        "type": "Wedding", "guests": 250,
-        "date": "May 14, 2027", "venue": "Private Estate, Kfar Yona Area",
-        "notes": "Chuppah at sunset. Vegan and gluten-free menu options required.",
+        "type": "Wedding", "guests": 250, "date": "2027-05-14",
+        "notes": "Chuppah at sunset. Vegan and gluten-free menu options.",
     },
 ]
 
 # ================================================================
-#  DATA — module level (no race condition)
+#  DATA LOADING — from HF Dataset repo (not local file)
 # ================================================================
 
-def _to_list(val):
+def _safe_to_list(val) -> list:
+    """Parse a column value to list regardless of storage type."""
     if isinstance(val, list): return val
     if isinstance(val, str):
         try:
@@ -104,84 +119,223 @@ def _to_list(val):
         except Exception: return []
     return []
 
-print("⏳  Loading vendor data …")
+
+def _load_vendors() -> pd.DataFrame:
+    """
+    Load the vendor dataset from HF Dataset repo eliel2003/events.
+    Tries three approaches in order:
+      1. datasets.load_dataset  (handles private repos via HF_TOKEN)
+      2. hf_hub_download        (direct file download)
+      3. pd.read_csv via URL    (public repo fallback)
+    """
+    token = HF_TOKEN or None
+
+    # ── Approach 1: datasets library ─────────────────────────
+    try:
+        from datasets import load_dataset
+        print("⏳  Trying datasets.load_dataset …")
+        ds = load_dataset(HF_DATASET, token=token)
+
+        # Find the vendors split — try common names
+        vendor_split = None
+        for name in ["vendors", "dataset_b_vendors", "vendor", "train"]:
+            if name in ds:
+                vendor_split = name
+                break
+        if vendor_split is None:
+            vendor_split = list(ds.keys())[0]
+
+        df = ds[vendor_split].to_pandas()
+
+        # If the dataset has both events and vendors in one split,
+        # filter to vendor rows using the vendor_id column pattern
+        if "vendor_id" not in df.columns and "event_id" in df.columns:
+            raise ValueError("Split contains events, not vendors.")
+
+        print(f"✅  Loaded {len(df):,} vendors from '{vendor_split}' split.")
+        return df
+
+    except Exception as e1:
+        print(f"⚠️  datasets.load_dataset failed: {e1}")
+
+    # ── Approach 2: hf_hub_download ───────────────────────────
+    try:
+        from huggingface_hub import hf_hub_download
+        print("⏳  Trying hf_hub_download …")
+        for fname in ["dataset_b_vendors.csv", "vendors.csv",
+                      "data/dataset_b_vendors.csv"]:
+            try:
+                path = hf_hub_download(
+                    repo_id=HF_DATASET, filename=fname,
+                    repo_type="dataset", token=token,
+                )
+                df = pd.read_csv(path)
+                print(f"✅  Loaded {len(df):,} vendors from '{fname}'.")
+                return df
+            except Exception:
+                continue
+    except Exception as e2:
+        print(f"⚠️  hf_hub_download failed: {e2}")
+
+    # ── Approach 3: direct URL ────────────────────────────────
+    print("⏳  Trying direct CSV URL …")
+    base = f"https://huggingface.co/datasets/{HF_DATASET}/resolve/main"
+    for fname in ["dataset_b_vendors.csv", "vendors.csv"]:
+        try:
+            headers = {}
+            if token:
+                headers["Authorization"] = f"Bearer {token}"
+            import urllib.request
+            req = urllib.request.Request(f"{base}/{fname}", headers=headers)
+            with urllib.request.urlopen(req, timeout=30) as r:
+                df = pd.read_csv(io.BytesIO(r.read()))
+            print(f"✅  Loaded {len(df):,} vendors via URL '{fname}'.")
+            return df
+        except Exception:
+            continue
+
+    raise RuntimeError(
+        f"Could not load vendor data from '{HF_DATASET}'. "
+        "Make sure the repository is public or set HF_TOKEN as a Space Secret."
+    )
+
+
+def _engineer_features(df: pd.DataFrame) -> pd.DataFrame:
+    """Apply the exact same feature engineering as EDA Cell 3."""
+    JSON_COLS = ["coverage_cities", "seasonal_availability",
+                 "specializations", "certifications"]
+
+    # Parse JSON list columns — exclude them from the str.strip() loop
+    for col in JSON_COLS:
+        df[col] = df[col].apply(_safe_to_list)
+
+    # Strip whitespace from plain string columns only (not JSON lists)
+    for col in df.select_dtypes(include="object").columns:
+        if col not in JSON_COLS and col != "vendor_profile_text":
+            df[col] = df[col].str.strip()
+
+    # Strip LLM artifact prefix from profile text
+    artifact = "**Vendor Profile:**"
+    df["vendor_profile_text"] = (
+        df["vendor_profile_text"].astype(str).str.strip()
+        .str.removeprefix(artifact).str.strip()
+    )
+
+    # Numeric features
+    df["day_rate_mid"] = (df["day_rate_min_usd"] + df["day_rate_max_usd"]) / 2
+
+    # Composite vendor quality score (mirrors EDA Cell 3 exactly)
+    r_min, r_max = df["avg_rating"].min(), df["avg_rating"].max()
+    df["rating_norm"]     = (df["avg_rating"] - r_min) / (r_max - r_min + 1e-9)
+    df["value_score"]     = 1 - (df["price_tier"] - 1) / 4
+    df["composite_score"] = (
+        0.4 * df["rating_norm"]
+        + 0.4 * df["sla_compliance_rate"]
+        + 0.2 * df["value_score"]
+    )
+    return df
+
+
+# ── Load and prepare data ─────────────────────────────────────
+print("⏳  Loading vendor data from HF Dataset repo …")
 try:
-    _df = pd.read_csv("dataset_b_vendors.csv")
-    for c in ["coverage_cities", "seasonal_availability",
-              "specializations", "certifications"]:
-        _df[c] = _df[c].apply(_to_list)
-    _df["day_rate_mid"] = (_df["day_rate_min_usd"] + _df["day_rate_max_usd"]) / 2
-    rmin, rmax = _df["avg_rating"].min(), _df["avg_rating"].max()
-    _df["rating_norm"]     = (_df["avg_rating"] - rmin) / (rmax - rmin + 1e-9)
-    _df["value_score"]     = 1 - (_df["price_tier"] - 1) / 4
-    _df["composite_score"] = (0.4 * _df["rating_norm"]
-                              + 0.4 * _df["sla_compliance_rate"]
-                              + 0.2 * _df["value_score"])
-    _VCITIES  = _df["coverage_cities"].tolist()
-    _VSEASONS = _df["seasonal_availability"].tolist()
+    _df = _load_vendors()
+    _df = _engineer_features(_df)
+
+    # Pre-extract arrays for vectorized filtering (Section 13 pattern)
+    _VCITIES  = [_safe_to_list(v) for v in _df["coverage_cities"]]
+    _VSEASONS = [_safe_to_list(v) for v in _df["seasonal_availability"]]
     _VCATS    = _df["category"].values
     _VRATES   = _df["day_rate_mid"].values
     _VCOMP    = _df["composite_score"].values
     _VIDX     = np.arange(len(_df))
-    print(f"✅  {len(_df):,} vendors loaded.")
-except FileNotFoundError:
-    print("❌  dataset_b_vendors.csv not found.")
+    print(f"✅  {len(_df):,} vendors ready.")
+
+except Exception as e:
+    print(f"❌  Vendor data load failed: {e}")
     _df = None
 
 # ================================================================
-#  EMBEDDING MODEL — CPU
+#  EMBEDDING MODEL — loaded from HF model repo
 # ================================================================
 
-print("⏳  Loading embedding model …")
+print(f"⏳  Loading embedding model ({EMBED_MODEL_ID}) …")
 _embed = SentenceTransformer(EMBED_MODEL_ID, device="cpu")
 
 if _df is not None:
-    print("⏳  Computing vendor embeddings …")
+    print("⏳  Encoding vendor profiles …")
     _vemb = _embed.encode(
         _df["vendor_profile_text"].tolist(),
         batch_size=128, show_progress_bar=True,
         normalize_embeddings=True, convert_to_tensor=True,
         device="cpu",
     )
-    print(f"✅  Embeddings: {_vemb.shape}")
+    print(f"✅  Embeddings ready: {_vemb.shape}")
 else:
     _vemb = None
 
-
-
-
 # ================================================================
-#  CORE ENGINE
+#  RECOMMENDATION ENGINE
+#  Scoring: 60% semantic similarity + 40% composite quality score
+#  (mirrors the design choice documented in Section 13 notebook)
 # ================================================================
 
-def recommend_vendors(brief, city, season, budget, top_n=3):
+def recommend_vendors(
+    event_brief:      str,
+    event_city:       str,
+    event_season:     str,
+    total_budget_usd: float,
+    top_n:            int = 3,
+) -> dict:
+    """
+    Stage 1 — Vectorized hard filters:
+      • City    : vendor must cover event_city
+      • Season  : vendor must be available in event_season
+      • Budget  : vendor day_rate_mid ≤ category-specific allocation
+
+    Stage 2 — Semantic ranking (60/40 blend):
+      final_score = 0.6 × cosine_similarity + 0.4 × composite_score
+
+    Returns {category: [vendor_dicts]} or {"error": str}.
+    """
     if _df is None or _vemb is None:
-        return {"error": "Vendor data not loaded. Check dataset_b_vendors.csv."}
-    if not brief.strip():
+        return {"error": "Vendor data not loaded. Check Space logs."}
+    if not event_brief.strip():
         return {"error": "Please enter an event description."}
 
-    city_ok   = np.array([city   in c for c in _VCITIES], dtype=bool)
-    season_ok = np.array([season in s for s in _VSEASONS], dtype=bool)
-    alloc     = np.array([budget * ALLOC_RATIOS.get(cat, 0.10) for cat in _VCATS])
-    budget_ok = _VRATES <= alloc
+    # Stage 1: hard filters (vectorized — no apply())
+    city_ok   = np.array([event_city   in c for c in _VCITIES], dtype=bool)
+    season_ok = np.array([event_season in s for s in _VSEASONS], dtype=bool)
+    alloc_vec = np.array(
+        [total_budget_usd * ALLOC_RATIOS.get(cat, 0.10) for cat in _VCATS],
+        dtype=float,
+    )
+    budget_ok = _VRATES <= alloc_vec
     combined  = city_ok & season_ok & budget_ok
     pool_idx  = _VIDX[combined].tolist()
 
     if not pool_idx:
+        n_c, n_s, n_b = int(city_ok.sum()), int(season_ok.sum()), int(budget_ok.sum())
         return {"error": (
-            f"No vendors matched all filters  "
-            f"(city: {int(city_ok.sum())}, season: {int(season_ok.sum())}, "
-            f"budget: {int(budget_ok.sum())}, combined: 0).  "
-            f"Try a larger budget or a different city."
+            f"No vendors matched all three filters.\n"
+            f"  City '{event_city}': {n_c} vendors\n"
+            f"  Season '{event_season}': {n_s} vendors\n"
+            f"  Budget ${total_budget_usd:,.0f}: {n_b} vendors\n"
+            f"  Combined: 0 vendors\n\n"
+            f"Try increasing the budget or selecting a different city."
         )}
 
-    q    = _embed.encode(brief, convert_to_tensor=True,
-                         normalize_embeddings=True, device="cpu")
-    sims = st_util.cos_sim(q, _vemb[pool_idx])[0].cpu().numpy()
+    # Stage 2: semantic similarity
+    q_vec       = _embed.encode(
+        event_brief, convert_to_tensor=True,
+        normalize_embeddings=True, device="cpu",
+    )
+    pool_embeds = _vemb[pool_idx]
+    sims        = st_util.cos_sim(q_vec, pool_embeds)[0].cpu().numpy()
 
     pool = _df.iloc[pool_idx].copy().reset_index(drop=True)
     pool["similarity"]  = sims
-    pool["final_score"] = 0.8 * sims + 0.2 * _VCOMP[pool_idx]
+    pool["final_score"] = 0.6 * sims + 0.4 * _VCOMP[pool_idx]
 
     results = {}
     for cat in VENDOR_CATEGORIES:
@@ -194,36 +348,40 @@ def recommend_vendors(brief, city, season, budget, top_n=3):
             ]].to_dict("records")
     return results
 
+# ================================================================
+#  OUTPUT FORMATTER
+# ================================================================
 
-
-
-def _stars(r):
+def _stars(r: float) -> str:
     n = min(5, max(0, int(round(float(r)))))
     return "★" * n + "☆" * (5 - n)
 
 
-def _fmt_vendors(recs, budget):
+def _fmt_vendors(recs: dict, budget: float) -> str:
     if "error" in recs:
         return f"### ⚠️ No Results\n\n```\n{recs['error']}\n```"
+
     lines = []
     for cat in VENDOR_CATEGORIES:
         if cat not in recs: continue
         alloc    = budget * ALLOC_RATIOS[cat]
         cat_name = cat.replace("_", " ")
         lines.append(
-            f"### {CATEGORY_EMOJI[cat]}  {cat_name}  ·  Budget: ${alloc:,.0f}\n"
+            f"### {CATEGORY_EMOJI[cat]}  {cat_name}  "
+            f"·  Budget ceiling: ${alloc:,.0f}\n"
         )
         for i, v in enumerate(recs[cat], 1):
             sp = v.get("specializations", [])
             if isinstance(sp, str):
                 try: sp = json.loads(sp)
                 except: sp = []
-            sc = v.get("final_score", v.get("similarity", 0))
+            sc = v.get("final_score", 0)
             lines.append(
-                f"**#{i}  {v['vendor_name']}**\n"
+                f"**#{i}  {v['vendor_name']}**  \n"
                 f"{_stars(v.get('avg_rating', 0))}  ·  "
+                f"{v.get('sla_compliance_rate', 0):.0%} SLA  ·  "
                 f"${v.get('day_rate_mid', 0):,.0f}/day  ·  "
-                f"Match `{sc:.3f}`\n\n"
+                f"Score `{sc:.3f}`\n\n"
                 f"*{', '.join(sp[:2]) if sp else '—'}*\n"
             )
         lines.append("---\n")
@@ -235,18 +393,53 @@ def _fmt_vendors(recs, budget):
 
 def handle_submit(brief, city, season, budget, ev_type,
                   date_from, date_to, guests, notes):
-    date_range = date_from if date_from == date_to else f"{date_from} → {date_to}"
-    recs      = recommend_vendors(brief, city, season, float(budget))
+    recs = recommend_vendors(brief, city, season, float(budget))
     return _fmt_vendors(recs, float(budget))
 
 
+def _date_html(df="2026-10-15", dt="2026-10-15"):
+    """Generate HTML calendar date range picker styled to match the palette."""
+    label_css = (
+        "font-size:.88rem;font-weight:500;color:#5C3D1E;"
+        "text-transform:uppercase;letter-spacing:.4px;"
+        "margin-bottom:6px;display:block;"
+    )
+    input_css = (
+        "width:100%;padding:9px 12px;border:1.5px solid #DDD0BE;"
+        "border-radius:10px;background:#fff;color:#2C1810;"
+        "font-family:Inter,sans-serif;font-size:.95rem;"
+        "box-sizing:border-box;cursor:pointer;"
+    )
+    sync_js = lambda eid: (
+        f"(function(v){{"
+        f"var el=document.querySelector('#{eid}');"
+        f"if(!el)return;"
+        f"var t=el.querySelector('textarea')||el.querySelector('input');"
+        f"if(t){{t.value=v;t.dispatchEvent(new Event('input',{{bubbles:true}}))}}"
+        f"}})(this.value)"
+    )
+    return f"""
+<div style="display:flex;gap:16px;margin:4px 0 12px;">
+  <div style="flex:1;">
+    <span style="{label_css}">Event Start Date</span>
+    <input type="date" id="ps_df" value="{df}"
+           style="{input_css}" oninput="{sync_js('ps_df_hid')}">
+  </div>
+  <div style="flex:1;">
+    <span style="{label_css}">Event End Date</span>
+    <input type="date" id="ps_dt" value="{dt}"
+           style="{input_css}" oninput="{sync_js('ps_dt_hid')}">
+  </div>
+</div>
+"""
+
+
 def _qs(idx):
     q  = QUICK_STARTERS[idx]
     b, c, s, bu = q["brief"], q["city"], q["season"], q["budget"]
     et, dt      = q["type"], q["date"]
     gs, nt      = q["guests"], q["notes"]
     vm = handle_submit(b, c, s, bu, et, dt, dt, gs, nt)
-    # Also update the visual date picker HTML so it reflects the quick starter date
     return b, c, s, bu, et, dt, dt, gs, nt, _date_html(dt, dt), vm
 
 def _qs0(): return _qs(0)
@@ -255,11 +448,12 @@ def _qs2(): return _qs(2)
 def _qs3(): return _qs(3)
 
 # ================================================================
-#  CSS
+#  CSS — WARM BROWN / CREAM / BEIGE PALETTE
 # ================================================================
 
 CSS = """
 @import url('https://fonts.googleapis.com/css2?family=Playfair+Display:wght@400;600;700&family=Inter:wght@300;400;500;600&display=swap');
+
 body, .gradio-container {
     background-color: #FAF7F2 !important;
     font-family: 'Inter', sans-serif !important;
@@ -267,14 +461,15 @@ body, .gradio-container {
 }
 .ps-header {
     background: linear-gradient(135deg, #3D2314 0%, #7A4E2D 60%, #B8895A 100%);
-    border-radius: 16px; padding: 36px 40px; margin-bottom: 20px;
-    box-shadow: 0 8px 32px rgba(61,35,20,.25); color: white;
+    border-radius: 16px; padding: 36px 40px; margin-bottom: 24px;
+    box-shadow: 0 8px 32px rgba(61,35,20,.25); text-align: center;
 }
 .ps-header h1 {
-    font-family: 'Playfair Display', serif;
-    font-size: 2.4rem; margin: 0 0 6px; letter-spacing: .5px;
+    font-family: 'Playfair Display', serif; font-size: 2.4rem;
+    font-weight: 700; color: #FAF7F2; margin: 0 0 6px; letter-spacing: .5px;
 }
 .ps-header p { color: #DDD0BE; font-size: 1.05rem; margin: 0; }
+
 label span, .label-wrap span {
     font-weight: 500 !important; font-size: .88rem !important;
     color: #5C3D1E !important; text-transform: uppercase !important;
@@ -283,16 +478,24 @@ label span, .label-wrap span {
 textarea, input[type="text"], input[type="number"] {
     background: #FFFFFF !important; border: 1.5px solid #DDD0BE !important;
     border-radius: 10px !important; color: #2C1810 !important;
+    font-family: 'Inter', sans-serif !important; font-size: .95rem !important;
 }
 textarea:focus, input:focus {
     border-color: #B8895A !important;
     box-shadow: 0 0 0 3px rgba(184,137,90,.12) !important;
 }
 input[type="range"] { accent-color: #B8895A !important; }
+
+.wrap-inner, .svelte-select {
+    background: #FFFFFF !important; border: 1.5px solid #DDD0BE !important;
+    border-radius: 10px !important; color: #2C1810 !important;
+}
+
 .qs-btn {
     background: #F5EFE6 !important; border: 1.5px solid #D4B896 !important;
-    color: #5C3D1E !important; border-radius: 10px !important;
-    font-weight: 500 !important; transition: all .2s !important;
+    color: #5C3D1E !important; font-family: 'Inter', sans-serif !important;
+    font-weight: 500 !important; border-radius: 10px !important;
+    padding: 10px 16px !important; transition: all .2s !important;
 }
 .qs-btn:hover {
     background: #EDE0CE !important; border-color: #B8895A !important;
@@ -300,74 +503,37 @@ input[type="range"] { accent-color: #B8895A !important; }
 }
 .submit-btn {
     background: linear-gradient(135deg, #5C3D1E 0%, #8B6239 100%) !important;
-    color: #FAF7F2 !important; font-size: 1.05rem !important;
-    font-weight: 600 !important; border: none !important;
-    border-radius: 12px !important; padding: 14px 28px !important;
-    width: 100% !important; margin-top: 8px !important;
+    color: #FAF7F2 !important; font-family: 'Inter', sans-serif !important;
+    font-size: 1.05rem !important; font-weight: 600 !important;
+    border: none !important; border-radius: 12px !important;
+    padding: 14px 28px !important; width: 100% !important;
+    margin-top: 8px !important;
     box-shadow: 0 4px 16px rgba(61,35,20,.25) !important;
 }
 .submit-btn:hover {
     background: linear-gradient(135deg, #3D2314 0%, #7A4E2D 100%) !important;
     transform: translateY(-1px) !important;
 }
-.status-bar {
-    background: #F5EFE6; border: 1px solid #DDD0BE; border-radius: 8px;
-    padding: 8px 14px; font-size: .82rem; color: #7A4E2D; margin-bottom: 14px;
+.prose, .markdown-body {
+    font-family: 'Inter', sans-serif !important;
+    color: #2C1810 !important; line-height: 1.7 !important;
+}
+.prose h3 {
+    font-family: 'Playfair Display', serif !important;
+    color: #5C3D1E !important;
+    border-bottom: 1px solid #DDD0BE; padding-bottom: 4px;
+}
+.prose hr  { border-color: #EDE0CE !important; }
+.prose code {
+    background: #F5EFE6 !important; color: #7A4E2D !important;
+    border-radius: 4px !important; padding: 1px 5px !important;
 }
-.prose, .markdown-body { font-family: 'Inter', sans-serif !important;
-    color: #2C1810 !important; line-height: 1.7 !important; }
-.prose h3 { color: #5C3D1E !important;
-    border-bottom: 1px solid #DDD0BE; padding-bottom: 4px; }
-.prose code { background: #F5EFE6 !important; color: #7A4E2D !important;
-    border-radius: 4px !important; padding: 1px 5px !important; }
 .ps-footer {
     text-align: center; color: #A68B6A; font-size: .78rem;
     margin-top: 28px; border-top: 1px solid #EDE0CE; padding-top: 14px;
 }
 """
 
-
-
-def _date_html(df="2026-10-15", dt="2026-10-15"):
-    """Generate HTML for the calendar date range picker."""
-    label_style = (
-        "font-size:.88rem;font-weight:500;color:#5C3D1E;"
-        "text-transform:uppercase;letter-spacing:.4px;margin-bottom:6px;display:block;"
-    )
-    input_style = (
-        "width:100%;padding:9px 12px;border:1.5px solid #DDD0BE;"
-        "border-radius:10px;background:#fff;color:#2C1810;"
-        "font-family:Inter,sans-serif;font-size:.95rem;"
-        "box-sizing:border-box;cursor:pointer;"
-        "accent-color:#B8895A;"
-    )
-    return f"""
-<div style="display:flex;gap:16px;margin:4px 0 8px;">
-  <div style="flex:1;">
-    <span style="{label_style}">Event Start Date</span>
-    <input type="date" id="ps_date_from" value="{df}"
-           style="{input_style}"
-           oninput="(function(v){{
-             var el=document.querySelector('#ps_df_hidden');
-             if(!el)return;
-             var ta=el.querySelector('textarea')||el.querySelector('input');
-             if(ta){{ta.value=v;ta.dispatchEvent(new Event('input',{{bubbles:true}}));}}
-           }})(this.value)">
-  </div>
-  <div style="flex:1;">
-    <span style="{label_style}">Event End Date</span>
-    <input type="date" id="ps_date_to" value="{dt}"
-           style="{input_style}"
-           oninput="(function(v){{
-             var el=document.querySelector('#ps_dt_hidden');
-             if(!el)return;
-             var ta=el.querySelector('textarea')||el.querySelector('input');
-             if(ta){{ta.value=v;ta.dispatchEvent(new Event('input',{{bubbles:true}}));}}
-           }})(this.value)">
-  </div>
-</div>
-"""
-
 # ================================================================
 #  UI
 # ================================================================
@@ -376,12 +542,13 @@ with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="ProSync AI") as demo:
 
     gr.HTML("""
     <div class="ps-header">
-      <h1 style="text-align:center;">ProSync AI</h1>
-      <p>The Event Producer's Command Center — intelligent vendor matching &amp; automated document generation</p>
+      <h1>ProSync AI</h1>
+      <p>The Event Producer's Command Center — intelligent vendor matching</p>
     </div>
     """)
 
-    gr.Markdown("#### ⚡ Quick Starters — click to auto-fill and run")
+    # ── Quick Starters ────────────────────────────────────────
+    gr.Markdown("#### ⚡ Quick Starters — click to auto-fill and search")
     with gr.Row():
         qs0 = gr.Button(QUICK_STARTERS[0]["label"], elem_classes=["qs-btn"])
         qs1 = gr.Button(QUICK_STARTERS[1]["label"], elem_classes=["qs-btn"])
@@ -391,56 +558,80 @@ with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="ProSync AI") as demo:
 
     gr.Markdown("---")
 
+    # ── Event inputs ─────────────────────────────────────────
     brief = gr.Textbox(
         label="Describe your event", lines=4,
-        placeholder="e.g. Tech summit for 400 guests — advanced AV, live streaming, "
-                    "kosher catering, VIP security…",
+        placeholder=(
+            "e.g. Tech summit for 400 guests — advanced AV, live streaming, "
+            "kosher catering, VIP security…"
+        ),
     )
     with gr.Row():
-        city   = gr.Dropdown(label="City",   choices=CITIES,   value="Tel Aviv",   allow_custom_value=False)
-        season = gr.Dropdown(label="Season", choices=SEASONS,  value="Summer",  allow_custom_value=False)
+        city   = gr.Dropdown(
+            label="City", choices=CITIES, value="Tel Aviv",
+            allow_custom_value=False,
+        )
+        season = gr.Dropdown(
+            label="Season", choices=SEASONS, value="Summer",
+            allow_custom_value=False,
+        )
+
     budget = gr.Number(
-        label="Total Budget (USD)",
-        value=200_000, minimum=5_000, maximum=2_000_000,
+        label="Total Budget (USD)", value=200_000,
+        minimum=5_000, maximum=2_000_000,
     )
 
     gr.Markdown("---")
 
+    # ── Document settings ─────────────────────────────────────
     with gr.Row():
         ev_type = gr.Dropdown(
             label="Event Type", choices=EVENT_TYPES, value="Tech Summit",
             allow_custom_value=False,
         )
-        guests = gr.Number(label="Guest Count", value=300, minimum=10, maximum=5000)
+        guests = gr.Number(
+            label="Guest Count", value=300, minimum=10, maximum=5000,
+        )
+
+    # Calendar date range picker (real <input type="date"> elements)
     date_picker = gr.HTML(value=_date_html())
-    date_from   = gr.Textbox(value="2026-10-15", visible=False, elem_id="ps_df_hidden")
-    date_to     = gr.Textbox(value="2026-10-15", visible=False, elem_id="ps_dt_hidden")
+    date_from   = gr.Textbox(value="2026-10-15", visible=False, elem_id="ps_df_hid")
+    date_to     = gr.Textbox(value="2026-10-15", visible=False, elem_id="ps_dt_hid")
+
     notes = gr.Textbox(
         label="Special Requirements",
-        placeholder="e.g. Kosher catering, black-tie dress code, vegan menu…",
+        placeholder="e.g. Kosher catering, black-tie dress code, outdoor setting…",
         lines=2,
     )
+
     submit = gr.Button(
-        "✨  Find Vendors & Generate Documents",
+        "🔍  Find Matching Vendors",
         elem_classes=["submit-btn"],
     )
 
     gr.Markdown("---")
+
+    # ── Results ───────────────────────────────────────────────
     gr.Markdown("### 🏪 Vendor Matches")
     vendor_out = gr.Markdown(
-        value="_Complete the form above to see recommendations._",
+        value="_Complete the form above and click **Find Matching Vendors**._",
         elem_classes=["prose"],
     )
 
-    gr.HTML('<div class="ps-footer">ProSync AI · Gradio + HuggingFace · '
-            'MiniLM-L6-v2</div>')
+    gr.HTML(
+        '<div class="ps-footer">'
+        'ProSync AI  ·  Gradio + HuggingFace  ·  '
+        'Dataset: eliel2003/events  ·  '
+        'Embedding: all-MiniLM-L6-v2  ·  '
+        'Scoring: 60% semantic + 40% quality'
+        '</div>'
+    )
 
-    # ── Wiring ────────────────────────────────────────────────
+    # ── Wiring ───────────────────────────────────────────────
     _in   = [brief, city, season, budget, ev_type, date_from, date_to, guests, notes]
     _out  = [vendor_out]
     _form = [brief, city, season, budget, ev_type, date_from, date_to, guests, notes]
-    _qs_out = [brief, city, season, budget, ev_type, date_from, date_to, guests, notes,
-               date_picker, vendor_out]
+    _qs_out = _form + [date_picker, vendor_out]
 
     submit.click(fn=handle_submit, inputs=_in, outputs=_out)
     qs0.click(fn=_qs0, outputs=_qs_out)
@@ -450,6 +641,4 @@ with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="ProSync AI") as demo:
 
 
 if __name__ == "__main__":
-    # server_name="0.0.0.0" is required on HF Spaces —
-    # without it Gradio raises ValueError: When localhost is not accessible
     demo.launch(server_name="0.0.0.0", server_port=7860)